Kazeko

So, she went under the "knife" this weekend, for a few changes and part additions - ram fans, and WAS going to try a different northbridge solution, but....



Turns out, nothing can fit on the southbridge that will fit under a super long graphics card like the 8800.

So, I picked up the ram fans (May not look amazing but it does the job) from microcenter (impatient, and the other fan alternatives I had were too large/bulky) and had tossed the north bridge idea.


I was with the thought that perhaps heat was the issue with the NB (MCP) but I can't find any alternatives so....



I took the advise of a guide on dh.net - "LAP EVERYTHING."

And away I go - I already lapped my TR 120 Extreme, but hadn't lapped much else (Thermalright SLK900A on my 2800+?) and I had to pull the whole system apart to install the fans over the ram, so what the heck.

Discovered a great many things.


1. Space is an issue with Thermalright Products, depending on the mainboard. - Previously I had my HR03 Plus on my 8800GTX wrapped around the card. This meant that the heatsink was almost right next to the Northbridge fan/heatsink, and the fan on the HR03+ couldn't fit all the way. - I've known this for a while and never had an issue with it...

But I opted to swap it's direction.

Result: 7C temperature drop on the GPU, 4C temperature drop on the CPU, 4C temperature drop on the Northbridge.


2. Space is STILL an issue. - Part of the reason no other northbridge solutions work on this board, is that there's no room to install them. I can't even imagine the space issue with two graphics cards.


3. The OEM EVGA 680i SLI north bridge cooler... Is brass.

BRASS.

I still lapped it, but I had to remove some of the "flair" from the heatsink, like the little "nVidia" logo plating on it (put it back on, but had to "modify it" to get it off/back on) but the fact that it's brass was what disturbed me.

4. So, my temps are now even lower, and the FSB still won't exceed 1400.

Perhaps I just got a dud? I'm not sure. It'll post @ 450x8 with the RAM 1:1 @ 4-4-4-12-1T - It appears stable until the whole system just locks out of the blue.

Sometimes sooner, sometimes later. There's no leaking caps, or appearance of physical damage, so I'm presuming I just got a dud somewhere along the line. I've altered every setting possible, and I think my solution at this point...


X38/48.


The mainboard's alright without overclocking, but it just cannot deliver very far beyond [Auto].



Pics!

Before
http://img86.imageshack.us/img86/1514/pict0021xz1.jpg


After
http://img126.imageshack.us/img126/5475/pict0028mq7.jpg


(Beer Sold Separately.)

(Not pictured: Another Aerocool Turbine on the rear)






Edit:


Some temps, after the change. Previously high 30's proc, high 40's GPU, and mid 30's North Bridge without load.

Now.
http://img523.imageshack.us/img523/3365/beforeot9.jpg


With some "Forced" cooling (read: turn off your heater and put a box fan at your door blowing cold air in)
http://img523.imageshack.us/img523/4836/afterqt0.jpg


I was kinda hoping for individual core temps in the single digits, but I'd have to really make my house chilly for that - I'd prefer to keep my plumbing.

CDsDontBurn

it'll post at 4-4-4-12 1t?!

at what voltages?

__________________

Mousey

Hah nice CPU Temps

__________________

Kazeko

2.2v - OE voltage

It's PC2-8500, dropped back to 450mhz from 533mhz.


And yes, it posts, and runs, but it locks whether the timings are that tight or not.


The more curious thing, is that it won't do 4-4-4-12-1t @ 400mhz, or 425. But it'll do it at 450, nothing beyond.

Hence my ardent quest for 450x8, so I can run 1:1

procupine14

wow sounds like you are having serious RAM issues similar to mine......i really need to get lucky and get a good set that i can actually OC

__________________
Laptop: HP dv6000 - Vista Ultimate (32-bit) - AMD Turion 64 X2 TL-60(2.0GHz) - NVIDIA GeForce Go 7200 256MB - 4GB Kingston DDR2 667 - 160GB 5400RPM SATA Hard Drive

HTPC: Vista Ultimate 64-bit - AMD Athlon 5200+ x2 - 4GB OCZ DDR2 800 - 250GB Seagate Barracuda - Samsung EcoGreen 1TB - Hauppauge single tuner HD - ASUS Radeon HD4550 512MB - ECS RS485M-M mobo - Creative X-fi soundcard

Other PC: Vista Ultimate 64-bit - AMD Athlon 4200+ X2 - 4GB Crucial Ballistix DDR2 800 - 250GB Seagate Barracuda - XFX 8600GT XXX - DFI Infinity NF570 SLI-M2/G

Kazeko

Why do you say that, if I may ask?

4-4-4-12-1t @ 450mhz vs. 5-5-5-15-2t @ 533mhz?

I'd say that's pretty amazing, given that I'm underclocking them so lightly.

It's not locking due to ram Timing issues. It's locking due to FSB issues (be it heat, or just unwilling silicon)

ChaosMinionX

1T or 2T is not a RAM specific thing in most cases, its a chipset/motherboard limitation, and it is usually based on NB voltage, not DRAM voltage....

And if you have to choose between 4-4-4-12 and 5-5-5-5-15 chances are 4-4-4-12 will bench better, and be faster despite the higher bandwidth because of the latency differences.

I dont usually choose 5-5-5-15 until about 1100mhz+ and just bump voltages accordingly.

__________________

procupine14

oh my bad i was confused.....no sleep etc lol

__________________
Laptop: HP dv6000 - Vista Ultimate (32-bit) - AMD Turion 64 X2 TL-60(2.0GHz) - NVIDIA GeForce Go 7200 256MB - 4GB Kingston DDR2 667 - 160GB 5400RPM SATA Hard Drive

HTPC: Vista Ultimate 64-bit - AMD Athlon 5200+ x2 - 4GB OCZ DDR2 800 - 250GB Seagate Barracuda - Samsung EcoGreen 1TB - Hauppauge single tuner HD - ASUS Radeon HD4550 512MB - ECS RS485M-M mobo - Creative X-fi soundcard

Other PC: Vista Ultimate 64-bit - AMD Athlon 4200+ X2 - 4GB Crucial Ballistix DDR2 800 - 250GB Seagate Barracuda - XFX 8600GT XXX - DFI Infinity NF570 SLI-M2/G

Kazeko

Okay so... I may have bested the beast.


At first I thought, N/B heat issue, bad board, what have you.

Do a little more homework and find out that there are a great many with my same board, same revision, running FSB speeds of 1600, 1700, 1800, and one @ 1900.

Further, I decided to grab a temp probe and start poking it around on the mainboard.


Temps are just dandy. NB (MCP) Is <40C @ 450x8.

Little more homework and apparently nTune = bad, even the temp monitor (which I was previously using.)

So, dump the monitor, and run Prime 95 by it's self - It doesn't die!

BUT! In order to keep it from dying, I couldn't say.... run Firefox AND P95. - Crashes if I do so.

I'm wondering if it's maybe mosfet heat from higher loads? (Considering throwing some AS5 on the heatsink/mosfets mebbe? Though I'd be concerned about potential conductivity and what not.)



So, I post here now, @ 450x8, temps are <55C on the CPU, <40C on the NB. RAM's 1:1 @ 4-4-4-12-1t.


It seems stable, until I throw something with a little more than average load at it - like P95 + Firefox. One or the other, and it has no qualms. I'll do some further stability testing with P95 - perhaps I can drop the vCore a bit, or narrow down the crashing problem further.


Edit: 2 hours on P95 with no interruption, Small FFT's. - The Larger FFT's with any other load (aside from Coretemp & CPU-Z) causes crashing, with 23C indoor ambient temps. - Leave the door open and drop the ambient temps in my house = No crash regardless. I'm pretty sure it's the MOSFET's.


Any other thoughts, or things I might be looking over? Seems temperature related, though all the board monitors are showing below 40c in most situations. (However accurate they may be.)


Edit2: So far, no errors or issues on P95 (though it has about a day's worth of testing left).

Big thing is... Memory Voltage = 2.125v vCore 1.39v @ 450x8

I'm hoping it'll hold this voltage for the next 24 hours or so. If it does, I'm gonna drop it down to 1.37 or 1.36 and see how it does.

Kazeko

This is why I've been so ardent about 450x8. I wants 4-4-4-12-1t.



Edit: Oh, where did you pick up another set of fan clips for your TR120, Chaos? I can't seem to find any place that sells them.

ChaosMinionX

Thermalright Flange Fan Wire Clips for Ultra-120 Heatsinks

Gary has them

__________________

Kazeko

Awesome, Thank you Sir.


Update: 10 hours stable.

Next Tests while @ work! Also, inc Thermalright HR09S/U's for the MOSFETs. - I'm almost positive that's the issue.


Update again @ 5:49 - It locked again after 1 hour of the Large FFT test in P95, but never had a rounding error or anything of that nature, so I'm going to be slowly dropping the voltage with the small FFT tests until I reach an acceptable limit, then I'll start testing with orthos and increase as necessary.


So... WOOTS. Me > 680i SLI, and no need for a Maximus Formula. (Unless I want Penryn/DDR3/Crossfire.)

Kazeko

I think power draw is the issue here, but it's really tough to tell. Low power draw (and the proper voltage) keeps it from crashing under low-load situations, but there is almost NO flexibility. I'm wondering if it's something else besides the mosfets that's causing the problem.

May order two sets of the HR09's just in case.

ChaosMinionX

Umm... if you going to be doing Orthos/Prime95 in Vista do the following...

Run/Services.msc then stop Superfetch service

Switch to classic windows theme, and disable the ability of the monitor to power off, turn it off yourself.


Superfetch messes with Prime95 hardcore, so just disable it while you stress.

Dont forget Priority 9, round off checking, and then run blend

__________________

Kazeko

Well, right, but it locks when using multiple progs, like a game AND Firefox - the more I load the CPU, the more it's likely to crash, sometimes with just one program up in a high(er) load situation. (I.e. Crashed with NWN2 at a specific spot, with no other programs running, superfetch off, classic theme.)

There's a set (of mosfets?) that get super hot right near the proc, labelled "R50" (yeah, I'm really not THAT deep with computers) right near the mosfet heat sink that i"m wondering about.

I.e. Gets too hot, voltage fluctuates too much or amperage goes fun-house mirror on me.

With or without Vista Aero/Monitor power off or Superfetch, it dies. It's purely load-based, and pretty sensitive. Hence why I"m thinking the Mosfets are just getting too hot or what have you. It almost seems to be based on raw power draw.

PSU can't handle the power draw maybe? (Though I wouldn't think it'd be THAT much draw with what I have.)

ChaosMinionX

Well... it sounds like your RAM isnt stable. Run blend and see what it does, large FFT and small FFT test CPU and whatnot more.

__________________

Kazeko

Blend = Locks within an hour.
Large FFT = Locks within an hour.
Small FFT = Fine, has gone 10+ hours.


RAM voltages varying everywhere from 2.125v to 2.25v. Locks up just the same. FSB voltage changed, doesn't change at all.


Edit: Found a few posts on the evga boards, and one on xtremesystems.org about improper mosfet cooling and memory voltage fluctuations because of it.

Looks like i"m holding stock until then.

Kazeko

Tried swapping the TIM on the mosfet coolers, still locks under heavy load. Doesn't even get through one series of tests in P95. Small it loves, has no issues with it.


I can sit here on 450x8 for hours if I want, until I forcefully load it down heavily (even runs 3dmark06 with no problems!)


It's almost like any more wattage than low-end use = crash.

I'm lost. Only have one other idea up my sleeve is to try and make HR09's fit properly (They're not supposed to, apparently) or throw a few RAM heatsinks on the "R50's" right by the mosfets.

If it is a RAM stability issue, I'm not sure how to correct it. Altering the timings changes nothing, and it's technically underclocked atm (normally 533) - 1T/2T makes no difference either. I can even clock it up to stock with the CPU @ 450x8, and it still locks.

Maybe I will be in the market for a new board.

ChaosMinionX

What are your current settings? I will be able to tell more from that

__________________

Kazeko

450x8 @ 1.38v (P95 tested stable*)
RAM - 1:1 With the CPU, @ 4-4-4-12-1T 2.2v
FSB - 1.5v (1800mhz)
SPP - 1.55v
MCP - 1.5v
SPP <-> MCP - 1.25v

All the appropriate options are disabled where necessary.



*I've had the vCore up to 1.55v, and it's more likely to lock at higher voltages. - Seems very wattage based, the higher the wattage draw, the more likely it is to die.

I've tested the ram from 400mhz, to 533, timings loosened/tightened from 4-4-4-12-1T to 8-8-8-20-2T at any given clock.

The only constant is if the FSB exceeds 1400mhz. - I can do 3.3ghz @ a 1477 FSB, with the ram @ stock clocks (1066, 5-5-5-15-2T) and it'll still lock, regardless of the CPU voltage.


Bad board mebbe? Unwilling Silicon? "Give up you moron and go buy a Maximus Formula?"




Also... thanks for you help Chaos. It comes greatly appreciated. DH.net, yourself and others have gotten me on this trail of OC'ing madness, which has been quite the enjoyable challenge. Thank you in advance.

ChaosMinionX

Ermm forget I asked settings, SPP and MCP are the NB and the thing for SLI I am not entirely sure of the voltage settings. I need to go read up on Nvidia mobo BS settings.... In all honesty I would say since your not SLI to go X38

__________________

Kazeko

Yeah... I'm beginning to think that. But it's just too much fun to toy with this...


Plus, the goal was to eventually get a second 8800GTX. But I may just go with an X38 mainboard anyway.

PangingJr

i would keep toying with it a little more myself before i give it up. it's probably not a cooling issue that causes the lock up. however, i have a little idea about your motherboard and your chipset voltage settings, but a quick search tells me to try adding more volts to the MCP, SPP and CPU's Core voltage, try leaving the FSB at 1.5 then the MCP 1.60, SPP 1.50 and Vcore to 1.50V-1.525V, also try running your DDR2 memory and CPU FSB @ 1:1 4-4-4-12 -2T 2.2V. then if this helps stop the lock up then try 4-4-4-12 -1T, and then later try lowering your Vcore and chipset's voltage...

Kazeko

Bumped the MCP voltage to 1.6v from 1.5v, SPP was already @ 1.5, FSB @ 1.5, RAM bumped to 2.2v, and the mem's @ 1T.


It'll be testing all night. I'll report back in the AM

ChaosMinionX

Hope it works out for you man I will do some reading and see what I can dig up about Nvidia boards beyond the normal Intel Settings.

__________________

PangingJr

here was the quick googling i did earlier... q6600 3.6GHz evga 680i
q6600 3.6GHz evga 680i - Google Search

then i checked out a few search results... the third ones at the SLI Zone Forums, Soulketcher680iSLI has posted...

Vcore 1.60
fsb: 1.5
MCP: 1.65
SPP: 1.55
HT: 1.5

just an idea of all the voltage uses on the board's and the processor

Kazeko

Well.

It ran the blend test for about 2 hours before it just locked again, last night.


No voltage set up I could run would allow 450x8, 1:1 on the RAM, 1T/2T didn't matter.



Funny thing is - 400x9 works just fine, with the RAM @ 1066 (4-4-4-12-2T) thanks to the MCP/SPP/HT voltage changes.


Downside is - the CPU wants more voltage @ the higher multi, and subsequently runs hotter.

Atleast it doesn't crash I suppose.

Gonna run a few tests to see how it performs (Yay trial version of everest until x-mas : / ) compared to the 450x8 (when it was stable.)



I seem to have picked up some talk of "FSB holes" on the EVGA forums, and SLI zone, after taking a bit more of a look around. - Perhaps that's the issue at present.


Edit: After playing around with this a bit more...


I get break-neck RAM speeds @ these "Hole" FSB clocks - It locks @ 369, but not 400 or 333.

Oddly enough, @ 3.3ghz (369x9), RAM @ 1066, 4-4-4-12-2T - I'm seeing read/write bandwidth in the area of 10gb/s, and sub .5ns latencies. (.46ns)

450x8, 1:1 - 4-4-4-12-1T provided for 9986mb/s read/write, and .53ns latency.

But.... those amazing RAM benchmarks are on FSB clocks that cause a lockup. - I'd think it the memory, but the 369x9 locks with the memory @ stock clocks.

So... 266/333/400? - All functional. - Outside that realm - Lockup.

ChaosMinionX

After re-reading this entire thread... I would say it has to do with the 680i's very limited QuadCore FSB capabilities, and would just leave it at 400x9, or go get an X38 or wait for the next highend chipset, I forgot that 680i even with new revisions was hit or miss with quads.

And if its a FSB hole as people were stating, try doing like 451x8 and see what that does, and bump it up until its stable, that G0 should be able to handle it.

__________________

Kazeko

Yeah, I think at this point, I'll be moving to a different mainboard.

Likely an X38 chipset, prolly a maximus formula. The big question is whether or not to hold out for DDR3 to become more standard, or just go with the DDR2 edition.

45nm support + PCIE2.0... the nForce700 series will carry it as well, but I've been a little.. "Disenchanted" with what I've heard about the MCP72IXE chipset.

It's the same Northbridge and Southbridge as the 680i SLI chipset, with an additional bus controller(s) for tri or quad SLI.

-- Overkill. If I was on a 40+ inch monitor, I might care more. - Even then CrossfireX is probably enough.


Tri and quad SLI, to me, is a method of pushing end-sales for manufacturers. - There's no real purpose beyond saying "OMG I HAS FOUR 8800 ULTRAS (or the upcoming G92 based cards)."

- I don't think there is, even on a benchmarking basis.



Plus. I miss ATI.

Edit: Thanks again, Panging, Virus, ChaosminionX, and everyone, for you help on this. Much appreciated!

ChaosMinionX

3870 x2 might be something to look at

__________________